Research On Dynamic Detection Technology Of Molten Silicon Liquid Level Based On Image Processing

Monocrystalline silicon is an important basic material in the contemporary information industry,and its growth process is highly valued by the industry and even the country.During the production of single-crystal silicon by the Czochralski method,the precise control of the molten silicon liquid level in the production process plays an important role in the quality of the single-crystal silicon,so it is of great significance to accurately detect the changes in the molten silicon liquid level.However,the Czochralski single crystal silicon growth furnace has the characteristics of high temperature(above 1000 °C),closed and negative pressure(about 15 Torr)environment,which makes it impossible to directly measure the molten silicon liquid level,so non-contact measurement is a common solution for molten silicon.In this paper,based on the natural structure reflection of the heat shield in the single crystal furnace on the molten silicon liquid level,research on the detection technology of molten silicon liquid level based on image processing was carried out.Firstly,after analyzing the advantages and disadvantages of the traditional methods,the composition of the molten silicon liquid level detection system was introduced.Then the mathematical model between the target features of the reflection image of the heat shield and the molten silicon liquid level was established,and some main hardware and software of the system were described.Then four experimental images were collected at the single crystal production site: calibration experiment,accuracy test experiment,crucible rotation experiment and crystal pulling process experiment.After comparing different image processing methods,an image processing algorithm based on the combination of local threshold segmentation and edge detection was proposed.Finally,the extracted feature value data was processed and analyzed.Calibration experiment and precision test experiment show that the change coefficient of the pixel area in image and the position of the molten silicon liquid surface is 270.1pixel / mm,and the detection accuracy is within 0.5 mm,which is more accurate than weighing method(centimeter-level accuracy).Crucible rotation experiment shows that the crucible rotation will cause deformation of the melt surface,the greater the crucible rotation speed,the greater the deformation of the molten silicon surface,and as the mass of the molten silicon decreases,the influence of the crucible rotation speed on the molten silicon level also increases.Crystal pulling process experiment shows that the fluctuation range of the molten silicon surface during the detection of equal diameter is about 1.5 mm,and the frequency is 0.36 Hz which is 3 times the rotation frequency of the ingot.The significance of the research work in this paper lies in that,on the one hand,through real-time accurate detection of the dynamic change of the molten silicon liquid position,the industrial control system can adjust the crucible position in real time according to the feedback information,maintaining the molten silicon liquid level at a relatively constant position,reducing the influence of latent heat during the phase transition of the crystal growth interface,and thereby improving the single crystal silicon production quality.On the other hand,measuring the changes of liquid surface vibration frequency and vibration amplitude through image processing is of great reference significance for analyzing the mutual coupling and action law between multiphysics fields of complex fluid systems.